Deteksi Sarkasme Pada Dataset News Headlines Menggunakan Artificial Neural Network Berbasis TF-IDF

Abstract

Sarcasm is a style of language frequently used in news headlines, especially in satirical media, thus posing a challenge for natural language processing systems in understanding the true meaning. The system's inability to recognize sarcasm can lead to misinterpretation in various tasks such as sentiment analysis and text classification. This study aims to detect sarcasm in news headlines using an Artificial Neural Network (ANN) with Term Frequency–Inverse Document Frequency (TF-IDF) feature representation. The dataset used is the News Headlines Dataset for Sarcasm Detection, which consists of approximately 28,000 news headlines from two sources, namely The Onion as sarcastic news and HuffPost as non-sarcastic news. The data is processed through a text pre-processing stage, then represented using TF-IDF with unigram and bigram schemes. The ANN model is trained using an 80:20 data split scheme and evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results show that the TF-IDF-based ANN model achieved an accuracy of 85.02% and an F1-score of 84.44% on the test data. These results demonstrate that the ANN approach with TF-IDF representation remains effective and efficient as a baseline method for detecting sarcasm in short texts such as news headlines.